Skip to main content
  1. Home
  2. Computing
  3. Features

AMD’s RDNA 5 might deliver what gamers want most — more VRAM

Add as a preferred source on Google
RX 7900 XTX installed in a test bench.
Jacob Roach / Digital Trends

Oh, alright then. Back on the AMD hype train I go. If you’ve had any interest in some of the best graphics cards in recent years, you probably know that the disturbing lack of video memory (VRAM) has been debated over and over. Games get more complex by the day, but not all GPUs are ready to face them for the years to come.

As the consumer market grows more and more critical of low VRAM on GPUs, it seems that AMD might be listening. A new leak shows upcoming RDNA 5 GPUs equipped with more VRAM than ever, and more juice to help them battle Nvidia.

AMD might boost VRAM across the board

Today’s leak comes from YouTuber Moore’s Law Is Dead (MLID), who is often a reputable source for all things PC hardware. Even still, take the following with a healthy dose of skepticism, as with any product that’s yet to launch, everything may still change.

Recommended Videos

With that disclaimer out of the way, MLID talked about AMD’s upcoming GPU lineup, also known as RDNA 5. These GPUs are still a long while away, but the YouTuber claims to have been given an internal document that talks about some of the models we might expect to launch between 2026 and 2027. Late 2026 is likely the earliest estimate for RDNA 5, given that the RDNA 4 lineup only arrived at CES 2025.

Moore’s Law Is Dead shared a watermarked version of the document that is said to come from a source at AMD. Right off the bat, we’re seeing big changes in the VRAM area.

For starters, AMD is said to be switching to GDDR7 VRAM with RDNA 5. This is great news, as GDDR7 memory is faster and offers higher bandwidth. Nvidia’s made the switch with the RTX 50-series, but AMD stuck to GDDR6 in its current generation of GPUs. It didn’t even upgrade to GDDR6X, which is slightly faster.

Now, we’re getting GDDR7 memory, and loads of it, too. The lowest amount of VRAM in a desktop gaming GPU is 12GB, which would likely end up in an RX 10060 XT (this is a placeholder name).

Not only that, but AMD is also reportedly going to use 36Gbps GDDR7 memory modules. That’s way faster than Nvidia’s current GDDR7, and nearly twice as fast as GDDR6.

The amount of VRAM goes up steadily, with the rumored RX 10070 GRE said to feature 15GB across a 160-bit bus (an unusual configuration). The RX 10070 XT could come with 18GB and a 192-bit bus, ramping things up.

Lastly, the RX 10090 XT seems to be the GPU where AMD delivers the big blow to Nvidia. Equipped with 36GB of VRAM across a 384-bit bus, it’d be AMD’s most powerful GPU to date.

More than just a memory boost

Beyond the VRAM gains, MLID is also showing some serious improvements in terms of compute units (CUs). He does, however, note that those are “incredibly subject to change,” so I’m not getting my hopes up … yet.

The flagship is said to sport 154 CUs — a massive uplift over the RX 9070 XT, which comes with 96. MLID speculates that the estimated equivalent GPU performance would be Nvidia’s RTX 6090, which is rumored to be the next-gen halo card from Team Green.

This kind of performance, and these kinds of specs, would put AMD firmly back on the high-end GPU market. Even if the RX 10090 XT (again, hopefully a placeholder name) ends up not quite beating Nvidia, it’ll still have a lot going for it, such as the 380-watt total board power (TBP). Nvidia’s RTX 5090 has a 575-watt TBP, so it’s easy to expect that the next-gen card will consume about the same. On the other hand, such a (comparatively) low TBP might mean that the card wouldn’t quite beat the RTX 6090.

Meanwhile, the lower-tier GPU would still be plenty powerful. Moore’s Law Is Dead says that the RX 10070 XT would offer the performance of an RTX 5080 or an RTX 4090, but at a mere $550 price point. I don’t know any gamers who wouldn’t love that.

What the GPU market desperately needs

Having lots of VRAM is crucial when gaming — especially at higher resolutions and at max settings. Complex, resource-heavy games like Cyberpunk 2077 munch on VRAM like it’s candy, and if they run out, it’s never a good time. Forcing the GPU to dip into system RAM ends in crashes or a massive performance loss.

This is why gamers and tech journalists alike have been rightfully complaining about the stagnant state of VRAM in GPUs. Mainstream cards like the RTX 5060 continue to launch with 8GB, which many believe is not enough for proper future-proofing. Of course, in reality, it all depends on what you expect to get from your graphics card — gamers who are content with 1080p at medium settings will still get use out of these cards for years to come. However, enthusiasts are right to worry.

Nvidia rules the GPU market, and with no pressure from AMD on the higher end, there’s very little reason for Nvidia to change its ways. If AMD can deliver a lineup of solid graphics cards, with plenty of VRAM and at a reasonable price, Nvidia might feel the heat and give us one or the other — better pricing or more VRAM. I’m not holding my breath for both.

Monica J. White
Monica is a computing writer at Digital Trends, focusing on PC hardware. Since joining the team in 2021, Monica has written…
Copilot could soon help diagnose issues with your PC
A new PC Insights feature will help you find what's slowing down your PC, though Copilot itself may be one of the main problems.
Microsoft Copilot Banner Featured

Copilot's next trick is diagnosing your PC's problems, but the catch is that the assistant doing the diagnosing is itself part of the problem. Windows Latest reports that Microsoft is testing a new Copilot feature called PC Insights, which will let you ask the AI assistant natural language questions about your computer's hardware and storage instead of digging through the Task Manager or Settings. The feature will reportedly allow users to ask questions like, "Do I have enough space for a 100GB game?" and Copilot will check the available storage to offer a response. Users will also be able to ask about CPU usage, battery health, etc., to diagnose issues.

What Copilot will be able to see

Read more
OpenAI just took the handcuffs off your ChatGPT Work and Codex usage limits, at least for now
The 5 hour limit is gone for now, your usage counter just reset, and GPT 5.6 Sol is about to get a lot lighter on your allowance.
OpenAI logo on Microsoft surface

It seems that OpenAI has been on a gallop lately, releasing new updates left and right. Just a few days back, the company launched its GPT-5.6 Sol, Terra, and Luna models for the general public, and now it's back with another update that will please its users even more. 

The headline change is that the 5-hour usage limit restriction in ChatGPT Work and Codex is being temporarily removed for all Plus, Business, and Pro plans. If you have ever been in the middle of a long working session and watched the limit message appear at the worst possible moment, this one is for you. 

Read more
This app gives your Mac a music player you’ll actually enjoy using
Apple Music on the Mac is a chore. Liqoria is the fix, and it plays nice with Spotify and YouTube too.
Liqoria music player

The Apple Music app on the Mac is not up to the mark. I don’t like how it looks or behaves, and Apple should take some inspiration from Spotify to make the app more modern and useful. Until that happens, we are stuck with a subpar app experience.

That’s why I never use the Apple Music app on my Mac and rely on third-party apps that let me control music. Today I am featuring one of the latest apps I discovered. It’s called Liqoria, and it’s probably the only music player app you will ever need.

Read more